KDnuggets

Mining insights in data science, machine learning, analytics, and AI • Edited by @mattmayo13 • KD stands for "knowledge discovery"

[KDnuggets @kdnuggets]

AI now appears in roughly one in five job postings, and agentic AI demand grew 587% in a year. Which AI skills employers actually name, by category and role. #Careers

https://t.co/DbSJAzI18p

[KDnuggets @kdnuggets]

Turn an existing Python function into an agent tool with one decorator. A website monitor becomes an OpenAI Agents SDK agent that decides what to check and why. #AIAgents

https://t.co/gyUAN6gIlF

[KDnuggets @kdnuggets]

We ran the open-source Humanizer model locally with llama.cpp and tested its rewrites against AI detectors. GPTZero's AI score dropped from 100% to 49% on our draft.

https://t.co/FjZzwZ4rC9

[KDnuggets @kdnuggets]

5 free courses that actually teach modern AI: DataCamp's AI for Work, the Easy-Vibe coding guide, Labonne's LLM Course, LLM Zoomcamp and Hugging Face's LLM Course. #LLM

https://t.co/lVkREJUYKz

[KDnuggets @kdnuggets]

Data can reveal the next move. Learn to turn insights into informed decisions with Georgetown’s 10-month M.S. in Business Analytics.
LEARN MORE
https://t.co/5uUNZ8KI2p

[KDnuggets @kdnuggets]

Stop picking local LLMs from leaderboards. Which of Phi-4-mini, Qwen2.5-Coder, Gemma 4, Llama 3.1 and others to run in Ollama for coding, reasoning, chat or RAG. #LocalLLM

https://t.co/zykhdOzPeR

[KDnuggets @kdnuggets]

Physical AI, Zero Trust, and NVIDIA Rubin: What's new at CoreWeave?
https://t.co/4yc4eRtE7I

[KDnuggets @kdnuggets]

Frozen embeddings plus a scikit-learn classifier vs. QLoRA fine-tuning on ag_news: 82% accuracy for the cheap route, more for LoRA at roughly 3x the time. When each one wins. #LLM

https://t.co/LEw6GUotJ3

[KDnuggets @kdnuggets]

5 tools for running agents in production, one per layer: LangGraph for logic, E2B for sandboxed code, Mem0 for memory, LangSmith for tracing and Modal for scale. #AIAgents

https://t.co/soECkaxkMp

[KDnuggets @kdnuggets]

One API, many models: use OpenRouter from Python to compare GPT, Claude and Llama, switch models with one string, and add fallbacks. Every example runs on free models.

https://t.co/kHcFKUEcvh

[KDnuggets @kdnuggets]

A 600M parameter model that runs offline on a laptop CPU and only cleans up transcripts, never rewriting what you said. What makes SuperWhisper's open-weight S1-mini different. #SpeechToText

https://t.co/fBM09LTG2I

[KDnuggets @kdnuggets]

Gemini Omni lets you edit a generated video turn by turn without redoing the scene. A clear look at its release timeline, model IDs, real costs, competitors and limits.

https://t.co/Uo0AXpJRrZ

[KDnuggets @kdnuggets]

5 free Microsoft courses on GitHub that take you from data science basics through scikit-learn and generative AI to AI agents, with lessons, quizzes and projects. #DataScience

https://t.co/NkhDQ5Ab5l

[KDnuggets @kdnuggets]

Python context managers for real cleanup: a timer that still reports on failure, TemporaryDirectory, patch.dict for reversible state, and ExitStack for resources you count at runtime. #Python

https://t.co/jaRG6GMnZs

[KDnuggets @kdnuggets]

Four stages from projective geometry to edge deployment, with Grounding DINO, SAM 2 and Depth Anything along the way, plus a portfolio project for each step. #ComputerVision

https://t.co/3ba89QvnAD

[KDnuggets @kdnuggets]

A JSON pipeline for ML configs that survives datetime, set, Path and Enum: a custom encoder, an object_hook decoder that fails gracefully, and a factory that migrates old schemas. #Python

https://t.co/TXhY4BB1ij

[KDnuggets @kdnuggets]

Do you need an AI agent or a workflow with an LLM step? One practical test and a five-point checklist to answer that before you write any code. #AIAgents

https://t.co/TCKUMwN3ic

[KDnuggets @kdnuggets]

7 Python design patterns for data science code, from Strategy for swappable metrics to Command for replayable runs as frozen dataclasses. Each comes with the caveat that keeps it honest. #Python

https://t.co/Lv12woUh0f

[KDnuggets @kdnuggets]

Multi-agent systems cost latency, tokens and debugging time. The four conditions where that complexity is worth it, and a framework for deciding. #AIAgents

https://t.co/LZ09XpycX1

[KDnuggets @kdnuggets]

Build a Python agent with the OpenAI Agents SDK that searches the web, calls your own functions as tools and remembers the conversation in SQLite, with no hand-written tool loop. #AIAgents

https://t.co/eRpmJ4bFTZ